6G Network Optimization: A Survey of OFDM-RIS Algorithms.

The optimization of joint Orthogonal Frequency-Division Multiplexing (OFDM) waveform design and Reconfigurable Intelligent Surface (RIS) configuration is a critical, complex challenge for 6G networks. This problem, which involves mixed-integer nonlinear programming, aims to maximize sum-rate, improve energy efficiency, ensure max-min fairness, and manage peak-to-average power ratio (PAPR). A comprehensive survey analyzed 78 research papers published between 2021 and 2026 on this topic. The survey categorizes existing optimization approaches into four main paradigms: model-based convex relaxation, heuristic and metaheuristic search, deep reinforcement and unsupervised learning, and emerging methods including foundation models, diffusion-based generative AI, and quantum optimization. A key finding from self-reported benchmarks is that machine learning (ML)-based methods (Paradigm III) achieve comparable spectral efficiency (95-99%) to model-based methods, but with significantly faster per-inference runtimes, often 100 to 10,000 times quicker, though this excludes ML pre-training costs. A companion tutorial benchmark revealed a crucial scaling property: GPU-based neural network inference runtimes remain constant regardless of problem size (N=16 vs. N=128), while iterative solvers scale polynomially. The survey identifies six major open challenges, including the absence of standardized cross-paradigm benchmarks, difficulties in real-world hardware deployment, multi-objective PAPR trade-offs, and the safety of large language models in live network control. It concludes by specifying requirements for a much-needed standardized benchmark to facilitate meaningful comparisons and accelerate research.